buffer: add indexOf() method

[trevnorris]

Add Buffer#indexOf(). Support strings, numbers and other Buffers. This
is more support than String#indexOf() gives, but the increased
versatility should be helpful.

Special thanks to Sam Rijs for first proposing this
change.

R=@bnoordhuis

This is a re-hash of #160. Done this way so future support can include regexp's and arrays.

[trevnorris]

Ref to original PR that brought this about: #160

[rvagg]

maybe else if, does that have any optimisation benefit here?

[trevnorris]

Good point. Will do.

[chrisdickinson]

We silently truncate values >255?

[trevnorris]

Am I supposed to throw?

[feross]

We already silently truncate values >255 when initializing a buffer:

new Buffer([256]) // <Buffer 00>

[chrisdickinson]

Mostly curious if we'd want to treat >0xff values as "search for byte sequence," though I suppose that could start to get into endianness issues.

[chrisdickinson]

(alternatively, yes, I would consider throwing for the sake of proper input validation.)

[trevnorris]

We're not going to throw. I was also considering the search for byte sequence. e.g. b.indexOf(0xbada55), and think it is a good solution. Though users will need to know that byte sequences over 0x20000000000000 loose precision. Also, I'm not worried about endianness. The input data is endiannes independent, and it's up to the user to know what the data looks like that they're searching for.

@bnoordhuis What would be the fastest way to convert a number to a byte sequence that can be searched for?

[bnoordhuis]

You mean like this?

uint32_t needle = args[1]->Uint32Value();
void* ptr = memmem(haystack, haystack_len, &needle, sizeof(needle));

[mscdex]

Missing ASSERT(args[2]->IsNumber()); ?

[trevnorris]

Hm. Guess I could put it in. The JS simply does coercion for sanity, but the Uint32Value() will still generally do the correct thing. Guess I'll throw it in.

[mscdex]

Ah ok.

[mscdex]

How would regexp functionality be implemented/supported? The built-in C regex library, v8's regex library, pcre, or?

[trevnorris]

@mscdex I'm working on that now. Can hack around it by having a pre-allocated external array class and reassign it on the fly, then convert the regex into 1 byte characters (for utf8 safety). So V8 would still do the heavy lifting but all the resources would still live outside the heap.

[rvagg]

see also #161 for lastIndexOf() (listing here for reference, not necessarily recommendation)

[bnoordhuis]

Suggestion: let's switch to std::min() and std::max() in a follow-up PR.

[feross]

@trevnorris Great work!

While considering how to add support in buffer, I realized I don't like how Buffer.prototype.indexOf behaves with a negative byteOffet value.

If byteOffset < 0, we currently search the whole buffer (same as passing 0). This is how String.prototype.indexOf behaves.

I think it makes more sense to treat a byteOffset < 0 as the offset from the end of the buffer. This is how Array.prototype.indexOf and TypedArray.prototype.indexOf do it. In general, I'd prefer to copy the way TypedArray works unless we have a good reason to do otherwise. Here's the spec.

[feross]

If we changed to Array.prototype.indexOf/TypedArray.prototype.indexOf semantics, this would change to:

assert.equal(b.indexOf('a', -1), -1);

And I'd add a few more tests for good measure:

assert.equal(b.indexOf('a', -4), -1);
assert.equal(b.indexOf('a', -5), 0);

[trevnorris]

thanks.